Docker 与 K8s 在性能测试中的应用

# 基础镜像：Eclipse Temurin JDK 17（LTS，性能优于 JDK 8）
FROM eclipse-temurin:17-jre-jammy

# 维护者与元数据
LABEL maintainer="perf-team" \
      org.opencontainers.image.title="jmeter-worker"

# 关闭代理，避免拉取插件时被拦截
ENV http_proxy="" https_proxy="" HTTP_PROXY="" HTTPS_PROXY=""

# JMeter 版本与路径变量
ENV JMETER_VERSION=5.6.3 \
    JMETER_HOME=/opt/apache-jmeter-5.6.3 \
    TZ=Asia/Shanghai
ENV PATH=${JMETER_HOME}/bin:${PATH}

# 安装必要的系统包并设置时区
RUN apt-get update && apt-get install -y --no-install-recommends \
        curl ca-certificates tzdata && \
    ln -snf /usr/share/zoneinfo/$TZ /etc/localtime && \
    echo $TZ > /etc/timezone && \
    rm -rf /var/lib/apt/lists/*

# 下载并解压 JMeter
RUN curl -fsSL \
    https://archive.apache.org/dist/jmeter/binaries/apache-jmeter-${JMETER_VERSION}.tgz \
    -o /tmp/jmeter.tgz && \
    tar -xzf /tmp/jmeter.tgz -C /opt && \
    rm /tmp/jmeter.tgz

# 安装常用插件：JMeter Plugins Manager + 自定义插件
RUN curl -fsSL \
    https://repo1.maven.org/maven2/kg/apc/jmeter-plugins-manager/1.10/jmeter-plugins-manager-1.10.jar \
    -o ${JMETER_HOME}/lib/ext/jmeter-plugins-manager.jar

# 压测脚本目录
VOLUME ["/scripts", "/results"]
WORKDIR ${JMETER_HOME}

# 非 root 用户运行
RUN useradd -m jmeter && chown -R jmeter:jmeter /opt /scripts /results
USER jmeter

# 入口：以 non-GUI 方式执行脚本
ENTRYPOINT ["jmeter", "-n", "-t", "/scripts/test.jmx", \
            "-l", "/results/result.jtl", \
            "-e", "-o", "/results/report"]

# k6：Grafana Labs 出品，原生支持云原生与 K8s
docker run --rm -i grafana/k6 run - <script.js

# Locust：基于 Python，分布式时通过 Master/Worker 模式编排
docker run --rm -p 8089:8089 -v $PWD:/mnt locustio/locust \
    -f /mnt/locustfile.py --headless -u 1000 -r 50

version: "3.9"

services:
  # JMeter Master：负责脚本分发与结果聚合，不实际产生压力
  jmeter-master:
    image: perf/jmeter:5.6.3
    container_name: jmeter-master
    volumes:
      - ./scripts:/scripts
      - ./results:/results
    command: >
      jmeter -n -t /scripts/test.jmx
      -R jmeter-worker-1,jmeter-worker-2,jmeter-worker-3
      -l /results/result.jtl -e -o /results/report
    depends_on: [influxdb]

  # JMeter Worker：实际产生压力，通过 -R 参数被 Master 调度
  jmeter-worker:
    image: perf/jmeter:5.6.3
    deploy:
      replicas: 3            # 横向扩展 3 个 Worker
      resources:
        limits:
          cpus: "2.0"        # CPU 上限，避免单 Worker 抢占宿主机
          memory: 2G
    volumes:
      - ./scripts:/scripts

  # InfluxDB v2：存储 JMeter Backend Listener 上报的指标
  influxdb:
    image: influxdb:2.7
    ports: ["8086:8086"]
    environment:
      DOCKER_INFLUXDB_INIT_MODE: setup
      DOCKER_INFLUXDB_INIT_USERNAME: admin
      DOCKER_INFLUXDB_INIT_PASSWORD: ${INFLUX_PWD:-perfPwd123}
      DOCKER_INFLUXDB_INIT_BUCKET: jmeter
      DOCKER_INFLUXDB_INIT_ORG: perf
    volumes:
      - influxdb-data:/var/lib/influxdb2

  # Grafana：可视化压测实时指标
  grafana:
    image: grafana/grafana:10.4
    ports: ["3000:3000"]
    environment:
      GF_SECURITY_ADMIN_PASSWORD: ${GRAFANA_PWD:-admin}
    depends_on: [influxdb]

volumes:
  influxdb-data:

docker stats --format "table {{.Name}}\t{{.CPUPerc}}\t{{.MemUsage}}\t{{.NetIO}}"

apiVersion: batch/v1
kind: Job
metadata:
  name: jmeter-load-test
  namespace: perf-test
spec:
  # 并行 Worker 数量，HPA 会动态调整
  parallelism: 5
  completions: 5
  backoffLimit: 2                # 失败重试次数
  activeDeadlineSeconds: 3600    # 任务最大执行时长，防止失控
  template:
    spec:
      restartPolicy: Never
      containers:
        - name: jmeter-worker
          image: perf/jmeter:5.6.3
          resources:
            requests: { cpu: "1", memory: "1Gi" }
            limits:   { cpu: "2", memory: "2Gi" }
          volumeMounts:
            - name: scripts
              mountPath: /scripts
              readOnly: true
            - name: results
              mountPath: /results
          env:
            - name: JMETER_WORKER
              valueFrom:
                fieldRef:
                  fieldPath: metadata.name
      volumes:
        - name: scripts
          configMap:
            name: jmeter-scripts      # 见 3.4 节
        - name: results
          emptyDir: {}

apiVersion: autoscaling/v2
kind: HorizontalPodAutoscaler
metadata:
  name: jmeter-worker-hpa
  namespace: perf-test
spec:
  scaleTargetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: jmeter-worker
  minReplicas: 2                    # 保底 2 个 Worker
  maxReplicas: 50                   # 上限 50 个，防止打爆集群
  metrics:
    - type: Resource
      resource:
        name: cpu
        target:
          type: Utilization
          averageUtilization: 70    # CPU 利用率超 70% 自动扩容
    - type: Pods
      pods:
        metric:
          name: jmeter_active_threads    # 自定义指标：活跃线程数
        target:
          type: AverageValue
          averageValue: 500

apiVersion: v1
kind: ConfigMap
metadata:
  name: jmeter-scripts
  namespace: perf-test
data:
  test.jmx: |
    <?xml version="1.0" encoding="UTF-8"?>
    <jmeterTestPlan version="1.2" properties="5.0">
      <!-- JMeter 脚本内容，由 CI 从仓库渲染 -->
    </jmeterTestPlan>
  users.csv: |
    user1,pass1
    user2,pass2

apiVersion: v1
kind: Namespace
metadata:
  name: perf-test
  labels:
    purpose: performance-testing
---
apiVersion: v1
kind: ResourceQuota
metadata:
  name: perf-test-quota
  namespace: perf-test
spec:
  hard:
    requests.cpu: "100"
    requests.memory: 200Gi
    pods: "100"
    # 限制压测任务最多占用 100 核 / 200G，防止吃满集群

apiVersion: argoproj.io/v1alpha1
kind: Workflow
metadata:
  name: perf-pipeline
  namespace: perf-test
spec:
  entrypoint: main
  arguments:
    parameters:
      - name: script
        value: "test.jmx"
      - name: duration
        value: "600"
  templates:
    - name: main
      steps:
        - - name: prepare-data
            template: prepare
        - - name: run-load
            template: jmeter-load
            arguments:
              parameters:
                - name: script
                  value: "{{workflow.parameters.script}}"
        - - name: report
            template: gen-report
            when: "{{steps.run-load.status}} == Succeeded"

    - name: prepare
      container:
        image: perf/data-prep:latest
        command: [sh, -c]
        args: ["python /app/prepare.py --output /tmp/users.csv"]

    - name: jmeter-load
      inputs:
        parameters:
          - name: script
      container:
        image: perf/jmeter:5.6.3
        resources:
          requests: { cpu: "4", memory: "4Gi" }
        command: [jmeter, -n, -t, "/scripts/{{inputs.parameters.script}}",
                  -l, /results/result.jtl, -e, -o, /results/report]

    - name: gen-report
      container:
        image: perf/reporter:latest
        command: [python, /app/report.py, --jtl, /results/result.jtl]

apiVersion: k6.io/v1alpha1
kind: K6
metadata:
  name: k6-sample
  namespace: perf-test
spec:
  runner:
    image: grafana/k6:0.51.0
    env:
      - name: K6_STATSD_ENABLE_TAGS
        value: "true"
  parallelism: 10                  # 10 个 runner Pod 并发
  script:
    configMap:
      name: k6-script
      file: test.js
  arguments: --out statsd --vus 1000 --duration 10m

# Prometheus 抓取 cAdvisor 的关键配置
scrape_configs:
  - job_name: cadvisor
    static_configs:
      - targets: ['cadvisor:8080']
    metric_relabel_configs:
      # 只保留 jmeter 容器指标，减少存储压力
      - source_labels: [container]
        regex: 'jmeter-.*'
        action: keep

# 实时观测压测 Worker 到被测系统的网络流
hubble observe --from-namespace perf-test --to-namespace prod \
    --protocol tcp --type flow

# 查看 TCP 重传，定位网络抖动
hubble observe --type trace --verdict DROPPED